Google's AI Agents Can Now Call Stores and Buy Things For You — Here's How

What Google Actually Launched

Google rolled this out in stages. The first wave landed in November 2025, when Google unveiled a suite of AI shopping upgrades inside Search's AI Mode and the Gemini app ahead of the holiday season. Vidhya Srinivasan, Google's VP and GM of Ads and Commerce, described it as an effort to strip out the "drudgery" still baked into online shopping — the endless scrolling, comparing, and phone-tag with local stores that shopping has always involved.

Three features made up that first wave:

• •Conversational shopping in AI Mode — you can describe what you want in plain language ("cozy sweaters for happy hour in warm autumn colors") and get back shoppable images, prices, reviews, and stock status, plus AI-generated comparison tables built from other shoppers' reviews
• •An AI that calls local stores for you — built on Google's Duplex technology (the same voice tech Google first demoed back in 2018), this feature calls nearby stores, asks about price, availability, and current promotions on a specific product, and reports back with a summary
• •Agentic checkout — an AI shopping agent that tracks a product's price against a target you set and automatically completes the purchase through Google Pay the moment the price drops into your range

By mid-2026, Google expanded the calling feature well beyond retail. At Google I/O in May 2026, Search VP Elizabeth Reid announced Google was extending "agentic booking" into local services — home repair, beauty, and pet care — meaning you can now ask Google to call a plumber or a salon on your behalf, not just check if a toy is in stock. Google said the feature would "roll out to everyone in the U.S." over the summer, running on Gemini 3.5 Flash as the default reasoning model for Google AI Pro and Ultra subscribers first.

How the Store-Calling Feature Actually Works

The flow is simpler than it sounds. You search for a product using "near me" or "nearby" phrasing in Google Search, and you're given the option to have Google confirm stock at a nearby location. Google then asks you a short set of questions tailored to that product — think size, color, or specific model — and asks whether you'd like the results sent by text, email, or both. From there, Google's AI places the actual phone calls to local stores, asks the questions on your behalf, and compiles a written summary comparing the responses.

For home services, the flow is nearly identical: the AI dials the business, asks the questions a homeowner would normally ask, and comes back with a summary that can end up comparing several competitors' pricing and availability side by side. The homeowner never has to speak to anyone directly — the call happens entirely between Google's AI and whoever answers the shop's phone, which increasingly is another AI system.

As of mid-2026, this feature is live in the U.S. for select categories, including toys, electronics, health and beauty products, and now home repair, beauty, and pet care services. It is not available everywhere — Google has excluded Indiana, Louisiana, Minnesota, Montana, and Nebraska from the rollout, and businesses can opt out entirely, either by contacting Google directly or adjusting settings in their Google Business Profile. Google also states the calls are monitored and recorded for quality assurance.

How Agentic Checkout Actually Pays For Things

The buying half of this is arguably the bigger deal, because it involves real money moving without you clicking "confirm" in the moment. To make that safe, Google built and open-sourced a new standard called the Agent Payments Protocol (AP2), announced in September 2025 with more than 60 launch partners, including Mastercard, PayPal, American Express, Coinbase, Adyen, Etsy, and Salesforce.

AP2 solves a specific trust problem: when an AI agent buys something on your behalf, how does a merchant or bank know the purchase was actually authorized by you, for that exact amount, and nothing more? AP2's answer is a system of cryptographically signed "Mandates," treated as verifiable digital credentials:

• •Intent Mandate — your original instruction to the agent ("buy these running shoes if they drop under $120," or "buy these concert tickets the second they go on sale")
• •Cart Mandate — created once the agent finds the specific item, capturing exactly what's being bought and at what price, before final authorization
• •Payment Mandate — authorizes the actual charge against a specific card, bank account, or stablecoin wallet, tied to the finalized cart

Because each mandate is signed and tamper-evident, it creates a non-repudiable audit trail — proof of what you asked for, what the agent selected, and what you were charged — for both the transactions where you're present and approving in real time, and the "delegated" transactions that happen while you're not even looking at your phone. AP2 is deliberately payment-agnostic, supporting traditional cards and bank transfers as well as stablecoins through an extension called x402.

Google frames agentic checkout as sitting on top of two things you already trust: its Shopping Graph (the product and pricing database behind Google Shopping) and Google Pay, which stores your payment credentials securely rather than handing them to the agent directly.

Google Isn't the Only One Doing This

This is part of a bigger industry shift happening in parallel. OpenAI and Stripe built a competing standard called the Agentic Commerce Protocol (ACP), which powered Instant Checkout inside ChatGPT starting in February 2026, initially with Etsy sellers, though OpenAI later scaled back in-chat purchasing in favor of an app-based model. Stripe and Tempo separately launched the Machine Payments Protocol (MPP) in March 2026, letting agents pre-authorize a spending limit and stream micropayments in both stablecoins and fiat currency. None of these protocols are mutually exclusive — a large retailer could plausibly support several at once — but Google's AP2 currently has the broadest coalition of payment and commerce partners behind it.

The Honest Limits

It's worth being clear about what this technology can't do yet. Industry coverage from mid-2026 points out that Google's calling feature remains narrow and shopping-shaped: calling four hardware stores to check if a part is in stock is a genuinely useful errand, but it's a small one. It is not yet capable of the calls people actually dread — rescheduling a dentist appointment, disputing a billing error, or sitting through a 40-minute airline hold queue. Those categories remain largely unserved by the major platforms for now, even as smaller AI-calling startups race to fill that gap.

There's also a second-order shift worth noting: as more stores start answering these AI-placed calls with their own AI systems, a growing share of "customer service" is becoming AI-to-AI conversation, with a human only reading the summary at the end.

Why This Matters

Google calling this "agentic checkout" isn't just a marketing label — it reflects a real architectural shift in how commerce works. Search, comparison, negotiation-by-phone-call, and payment used to require a human at every step. Google has now automated three of those four steps for a growing list of product categories, with the fourth — actually deciding what to buy — increasingly delegated too, as long as you're comfortable setting the rules in advance. Whether that counts as convenience or as one more layer between you and your own spending is likely to be the real debate as this rolls out further through 2026 and beyond.